Chaotic Optimization

Andrew Andrade (andrew@andrewandrade.ca)
January 4, 2015

Slides online: mrandrewandrade.com/talks/chaotic-optimization-jan-4-2016

The past determines the future

BUT

Approximate past did not approximately determine my future

Simple Function

f(past) = future

f(past+ δ) = new future

\[ \frac{f(past+ δ )}{f(past)} = \Delta future = ? \]

\[ \frac{f(past+ δ )}{f(past)} = \Delta future = ? \]

You can (randomly) change the future!

Proof by story!

Port Credit

Highschool Co-op

Machineshop!

Sweeping -> Programming CNC

1st co-op: Imtex membranes

Startup doing gas filtration

Plastic Injection Molding

Got the job because of machine shop experience

2nd co-op: Suncor (PetroCanada)

Reservoir engineering

3rd co-op: Peraso

Fabless Semiconductor

Ended up designing their first to market product

4th co-op: Facebook

Facebook Manufacturing!

Luck is

when preperation

meets opportunity!

Larry Smith:

Why you will fail to have a great career

Why you will fail

World of machine learning

5th co-op: PetroPredict

Global Petroleum Show

Equivilent Exchange

All decisions come with costs!

Life's a Rollercoaster

Oil price

Learning: Business Need Comes First!

Pivot: Pipeline Risk

Taken from: http://www.mero.cz/ (can't put real data due to NDA)

Running Servers

Re-presenting the talk at the Ontario Engineering Competition 2016

Canada wide SPE Paper Contest 2014

Third place in masters/undergrad!

news release

Singapore!

Self driving bicycle!

SE Asia!

What's Next

Capstone: Energy Storage

Follow the progress: mrandrewandrade.com/blog

Paperscraper

Early Results

Petroleum101.com

DataScienceGuide.github.io

Takeaways

School sucks…

xkcd.com/519/

Digital (Computer) Controls

Digital (computer) Controls

School is very important

Its ALL about applying what you learn

Example: Digit Underflow

\[ P(x_1) = 0.00000000001 = 1 \times 10^{-10} \]

\[ P(x_2) = 0.000000000000000000001 = 1 \times 10^{-20} \]

\[ P(x_3) = 1 \times 10^{-30} \] \[ P(x_4) = 1 \times 10^{-100} \] When where are too many digits, it takes lots of memory to save and causes problems

Digit Underflow Solution

Digit Underflow Solution

\[ - \log(P(x_1)) = - \log (1 \times 10^{-10}) = 10 \] \[ - \log(P(x_2)) = - \log (1 \times 10^{-20}) = 20 \] \[ - \log(P(x_3)) = - \log (1 \times 10^{-30}) = 30 \] \[ - \log(P(x_4)) = - \log (1 \times 10^{-100}) = 100 \]

Another example: Machine intelligence

Design a computer vision system for a copter to land in water

Solution: Decision Tree

More examples: Computer vision

\[ Split (T,X) = H = - \sum_i p_i \log_b p_i \]

\[ H( X \text{or} Y) = 2\Big(-{1\over 4} \log_2 {1\over 4}\Big) - {1\over 2} \log_2 {1\over 2} = 1.5 \]

Hackathons

Teams/Startups

Side projects

Side projects

Take away

It comes back, but even worse!

Resources

Khan Acedemy
Udacity
EdX
Cousera
…

Many more!

Do great things!

Questions?

Andrew Andrade

andrew@andrewandrade.ca

mrandrewandrade.com/talks/chaotic-optimization-jan-4-2016

Chris Thiele

christhiele@live.com